The Turing Valley: How AI Capabilities Shape Labor Income

Abstract

Current AI systems are significantly better than humans in some intelligence dimensions but remain weaker than humans in others. Guided by the long-standing vision of machine intelligence embodied by the Turing Test, AI developers increasingly seek to eliminate this "jagged" nature of AI by pursuing Artificial General Intelligence (AGI) with consistently high performance across domains. This pursuit has sparked a heated debate, with leading economists arguing that AGI risks eroding the value of human capital by turning machines into closer substitutes for labor. We contribute to this debate by characterizing how different paths of AI progress shape labor income in a multidimensional knowledge economy. We show that AI improvements in its strong dimensions always increase labor income, but the effects of AI progress in its weak dimensions fundamentally depends on the nature of human–AI communication. When communication is able to integrate partial solutions, AI improvements in its weak dimensions decrease labor income. In this case, as long as humans are sufficiently strong in one dimension, labor income is maximized by a deliberately jagged form of AI. In contrast, when communication is only able to share full solutions, AI improvements in its weak dimensions increase labor income provided that AI is sufficiently stronger than humans in its strong dimensions. In this case, if humans are sufficiently weak in one dimension, labor income is maximized when AI attains high performance across all dimensions. These results highlight the importance of empirically assessing the properties of human–AI communication for understanding the labor-market consequences of progress toward AGI.